CyberdeceptionAtlas · Question → testES

Library / Paper

Paper

A Partial Break of the Honeypots Defense to Catch Adversarial Attacks

According to the authors, this paper presents an attack that partially breaks a defense that injects 'honeypots' into neural networks to detect adversarial attacks. They report reducing the detection true positive rate to 0% and detection AUC to 0.02 while maintaining original distortion bounds, and note that the original defense authors later amended it to mitigate the attack.

HoneypotAdversary engagementApplication

What we know

According to the authors, this paper presents an attack that partially breaks a defense that injects 'honeypots' into neural networks to detect adversarial attacks. They report reducing the detection true positive rate to 0% and detection AUC to 0.02 while maintaining original distortion bounds, and note that the original defense authors later amended it to mitigate the attack.

Evidence limits

This entry is based solely on the publicly available title and abstract; the full text has not been reviewed nor has the attack's replication been independently verified.

Reviewed: author abstract. Full-text results were not analyzed.

Sources and provenance

  1. A Partial Break of the Honeypots Defense to Catch Adversarial Attacks
    abstract · 2026-10-04

Reviewed: 2026-10-04. This record may change when new evidence is found.

RIS · BibTeX